1: \begin{abstract}

2: 

3: We propose a framework, named the postselected inflation framework, to obtain converging outer approximations of the sets of probability distributions that are compatible with classical multi-network scenarios. 

4: Here, a network is a bilayer directed acyclic graph with a layer of sources of classical randomness, a layer of agents, and edges specifying the connectivity between the agents and the sources.

5: %

6: %, such that the edges of the graph are always between an agent and a source. 

7: A multi-network scenario is a list of such networks, together with a specification of subsets of agents using the same strategy.

8: %

9: %An outer approximation of the set of multi-network correlations provides means to certify the infeasibility of a list of agent outcome distributions.

10: %It also allows one to obtain upper bounds for maximization problems under such multi-network constraints.

11: %

12: %We present our results in a specialized tensor network notation, and we make our results explicit following the thread of a simple example which we call the Correlated Sleeper, in analogy to the setting of the Sleeping Beauty paradox.

13: %

14: %The algorithmic tests under considerations are linear programs.

15: %These tests will converge to the true answer in the limit of high-order tests, although the convergence is rather slow compared to the computational cost of these tests. The first low-order tests are however tractable and predictive.

16: %

17: %

18: We furthermore show that the postselected inflation framework is mathematically equivalent to the standard inflation framework: in that respect, our results allow to gain further insights into the convergence proof of the inflation hierarchy of Navascuès and Wolfe, and extend it to the case of multi-network scenarios.

19: 

20: 

21: \end{abstract}

22: